Protocol number 93M0170 The Section on Functional Imaging Methods (SFIM) has advanced functional MRI (fMRI) methodology through development of sophisticated processing and acquisition methods and research on the underlying mechanisms behind the fMRI signal. The ultimate goals are 1: a deeper understanding of healthy human brain and 2: clinical application of fMRI on an individual patient basis.
SFIM aims to increase the depth and breadth of fMRI applications and bridge the gap to clinical applications on individual patients. Multi-echo fMRI to improve connectivity mapping Carried out primarily my graduate student, Prantik Kundu, in collaboration with Dr. Edward Bullmore of Cambridge University: The present work built upon the novel methodology, developed by us, that combines NMR measurements from multi-echo fMRI (T2* signal decay analysis) with statistical decomposition (independent components analysis, ICA) to differentiate functionally related blood oxygenation level dependent (BOLD) signals from the noise. This method is called multi-echo ICA (ME-ICA). This years work was directed towards applying ME-ICA to enhance resting state connectivity mapping, implementing ME-ICA for multi-echo multi-slice fMRI at 3T, denoising and characterizing 11.7T rodent multi-echo fMRI at varying anesthesia, and completing our study that introduced a new approach for cortical connectomics enabled by ME-ICA decomposition. Detection of low frequency task signals in fMRI after multi-echo denoising Carried out by my post doc, Jen Evans, as well as Prantik Kundu and in collaboration with Silvina Horvitz of NINDS: Assessment of slow BOLD changes is critical in studies of slow drug effects, transcranial magnetic stimulation (TMS) induced changes and other treatments that may involve slow cognitive changes. Conventional single echo fMRI cannot separate non-BOLD based signal drifts from neuronally-related changes limiting the types of paradigms that can be used. In this project we demonstrate that multi-echo independent components analysis (ME-ICA) can separate these two mixed low frequency signals. Biological significance of wide-spread fMRI activations at high TSNR Carried out by my post doc Javier Gonzalez-Castillo and post bac IRTA, Colin Hoy: We have previously demonstrated that with adequate temporal signal-to-noise ratio (TSNR) and sufficiently versatile response models, statistically significant BOLD responses time-locked with the experimental paradigm can be detected in over 95% of the brain for simple tasks. We also showed that these widespread responses cluster spatially in a functionally meaningful manner. In order to better evaluate the biological significance of these widespread activations, we conducted an additional experiment in which task load was modulated across participants. Despite lower spatial smoothness, TSNR, and contrast-to-noise ratio (CNR) in the present dataset relative to our original study, still over 80% of gray matter became significantly active at the highest available TSNR. Activation extent scaled with task load and followed the gray matter contour, suggesting biological significance for activations found outside of cortical areas commonly associated with the tasks. These results fundamentally challenge how typical activation processing - using a simple canonical model for expected response - is performed. More information is clearly in the signal and we are effectively extracting much of it using this method. Cognitive correlates of dynamic changes in connectivity Carried out by my post doc Javier Gonzalez-Castillo and post bac IRTA, Colin Hoy: A common assumption in most resting state fMRI (rsfMRI) studies is temporal stationarity for the duration of the scan. However, recent studies have shown that rsfMRI spatial connectivity patterns do change considerably across short periods of time. The potential correlation between connectivity changes and ongoing cognitive processing is not fully understood. The purpose here was to evaluate whether dynamic changes in whole brain fMRI connectivity can be used to reliably infer cognitive state at the single-subject level using unsupervised methods. Subjects have been scanned in a 7T MRI scanner continuously for approx. 25 mins as they performed and transitioned between four distinct mental tasks: undirected rest, 2-back memory task, simple math, visual attention. Each task was performed for 3 mins on two different occasions within the 25 mins of scanning. Our results show that connectivity patterns contain sufficient information to correctly classify time-periods according to ongoing mental processes well above chance for window durations longer than 30s. In these experiments we found that combining multidimensional scaling (feature reduction) and k-means (classification algorithm) provide the best results. Hippocampal Variability across subject and with intervention. Carried out by my graduate student, Adam Thomas in collaboration with Dr. Heidi Johansen-Berg at the University of Oxford. The first project furthers the labs continued interest in longitudinal methods and within-subject designs by looking at the affect of aerobic exercise on the hippocampus. This is the first finding from a large intervention experiment in which we have shown a volume growth in hippocampus that is dominated by increases in myelination. As second arm of this study is the exploration of the relationship between Cerebral Blood Volume (CBV), aerobic exercise and fitness. Our data show correlations between CBV and fitness levels in humans. Another project is looking at high resolution, 7T hippocampal morphometry. With over 30 subjects scanned, we have uncovered a high degree of individual variability in hippocampal morphometry previously not visible at 3T. Some participants show a high degree of convolution in the hippocampal sheath, while others are relatively smooth. We are currently devising methods to quantify this convolution and correlate it against behavioral variables such as fitness and memory. In collaboration with Dr. Carlos Zarate, we have also scanned a large cohort of patients suffering from drug-resistant depression before and after the administration of ketamine.
We aim to explore whether the rapid behavioral effects of ketamine have any structural correlates in hippocampus. Finally, we are exploring a similar collaboration with Dr. Judy Rapoport to use the same sequence to study hippocampal morphometry in childhood-onset schizophrenia. Data-driven methodology for patient subtype discovery Carried out by my post doc, Zhi Yang. An obstacle in searching for neuroimaging biomarkers for mental disorders is that the existing definitions of patient groups are based on behavioral symptoms and may not reflect the pathophysiological characteristics of the disease. This fact leads to high heterogeneity and poor reproducibility in psychiatric studies. Potential pathophysiological subtypes can be overlooked in studies grouping patients based on clinical diagnosis. We are exploring a data-driven approach, gRAICAR, to search for highly homogeneous patient subgroups based on their intrinsic brain connectivity patterns. gRAICAR is unique in that it reveals inter-subject variability in brain connectivity patterns. Based on the brain network variability, patient subgroups suggesting potential pathophysiological subtypes can be identified using community detection algorithms. Two novel clinical findings demonstrated the power of gRAICAR: Using gRAICAR, we separated a precuneus network from the default mode network according to their different cross-lifespan trajectories. We also uncovered an association between a frontotemporal network and the severity of positive and negative symptoms in early-onset schizophrenia.
Showing the most recent 10 out of 84 publications
